1. Connect Slack
Install the workspace integration, pick a department, and connect the channels the swarm should watch.
Docs
Set up a swarm in Slack. Seed it once. Let it compound.
Sockt docs focus on activation, memory, and safe autonomy. The old SDK/MCP-for-payments framing has been retired.
2. Seed memory
Answer five setup questions so the swarm knows your ICP, offer, approval rules, and operating context.
3. Bring your own key
Connect Anthropic, OpenAI, Azure OpenAI, Google Gemini, or a self-hosted provider. Sockt never adds inference markup.
4. Activate
The swarms spin up automatically and begin monitoring channels, coordinating tasks, and writing back into memory.
Example workflows
Growth swarm: watches for buying intent, enriches leads, and drafts outbound for approval.
Content swarm: builds a weekly content calendar, writes X threads and LinkedIn posts, generates AI video (LTX draft clips → Kling 3.0 final → Shotstack stitch), submits everything to Slack for approval, then publishes to platform APIs.
Product swarm: turns a request into a structured spec, runs approved tasks, and delivers a consolidated PR.
Ops swarm: catches incidents, correlates history, and drafts the root-cause summary.
Operational guardrails
- Safe actions can run autonomously; risky actions default to review.
- Keys stay isolated from agent context windows.
- Memory is Git-backed so the team can audit what changed and why.
Open source
Read the code, not just the pitch.
The orchestrator, agent runtime, FSM, memory pipeline, and CLI are all open-core (FSL-1.1-MIT, MIT after 2 years). Full technical docs live in the repo.
Architecture →
How the packages fit together, the task FSM, the agent execution loop, the memory pipeline.
API Reference →
Full orchestrator HTTP API — tasks, agents, approvals, health.
Configuration →
Every environment variable, what reads it, and free-tier LLM tuning notes.
Departments & Skills →
The skill index pattern, the .skill file format, and how to add a department.
Contributing →
Dev setup, repo layout, conventions, and the PR process.
Security →
Vulnerability reporting and the current sandbox/security boundaries.